Efficient search for new physics using Active Learning in the ATLAS Experiment

Searches for new physics at the LHC set exclusion limits in multi-dimensional parameter spaces of various theories. Typically, these are presented as 1- or 2-dimensional parameter scans; however, the relevant theory's parameter space is usually of a higher dimension. As a result, only a subspace is covered, which is due to the computing time requirements of simulations for the signal process. An Active Learning approach is presented to address this limitation. Compared to the usual grid scan, it reduces the number of points in parameter space for which exclusion limits need to be determined. Hence it enables richer interpretations of searches in higher-dimensional parameter spaces, which increases the value of the search. For example, this may reveal regions of parameter space that are not excluded and motivate new, dedicated searches. Our Active Learning approach is an iterative procedure. First, a Gaussian Process is fit to exclude signal cross-sections. Within the region close to the exclusion contour predicted by the Gaussian Process, Poisson disc sampling is used to sample additional points in parameter space for which the cross-section limits should be evaluated. The procedure is aided by a warm-start phase based on computationally inexpensive, approximate limit estimates. A python package, excursion, provides the Gaussian Process routine. The procedure is applied to a dark matter search performed by the ATLAS experiment, extending its interpretation from a 2 to a 4-dimensional parameter space while keeping the computational effort at a low level.